stelliferous

Search uses release 0.0.2.0 · 2,086 definitions

Releases (2)

type Context
def window_size
def plan_size
def parallelBytes price loading, storing and the loop that moves them (copy_byte_ns is measured on copy loops); operations count the rest.
def pooledThe destination of the two sweeps of channels stored planes: along the columns into middle, [channels,height,out_width], then along the rows from origin into output.
def windowed
def leaf
def prepared
def planned
def execute

type Size
def merge
def size
def sizedThe whole operation, not each leaf, must reach the grain: small leaves only lengthen the runtime's rounds, which the region estimate already counts.
def parallel
def allocation_elementsFor equal-shaped operands the logical read/write volume of repartition is identical.
def alignment
def mapped_context
def map_context
def scalar_planned
def scalar
def indexed_planned
def indexed
def transformed
def transform

type Geometry
type Rows
type RowWindowA leaf's output rows, its halo rows above and below the core, and the image rows its window holds: from the stride-aligned first row its outputs read to the last, so the window packs as a shorter image.
def ceiling
def subtract_clipped
def clipped
def nonempty
def startThe first row a window holds: the first row its outputs read, moved down to a stride multiple.
def windowOutput partition boundaries follow the input cores, with the first and last cores also owning any output rows introduced by padding.
type Halo
def Halos
def extract
def halo
def copied_window
def assembledThe window rows of every channel, stored contiguously.
def assemble
def input_geometry
def prepareThe full core tree is threaded through the preparation pass without cloning.
def window_band
def applyEvery core leaf with its halo, through the shared band recursion.

type InputShape
type Geometry
type DenseArguments
def input_shape
def matched
def geometry
def output_shape
def selected_depth
def configured_depth
def depthThe input tensor's workers bound the split, as a thread pool bounds PyTorch's parallel_for.
def copied_output
def image
def batch
def images_selected
def images
def execute_dense_at_depth
def dense_arguments
def execute_dense_arguments
def execute_dense
def band_context
def band_work
def context_window
def band_output
def start_bands
def initial_layout_selected
def initial_layout
def execute_many
def uses_band_windowsPadding-only output windows and an empty batch use the existing dense path.
def execute_layout
def execute_selected
def execute
def reorderedThe weights in the order the kernel reads them: once, after the checks.